March 2024 - Wine-commits - list.winehq.org

Zebediah Figura : d3d11/tests: Test discarding a buffer in test_high_resource_count().
by Alexandre Julliard 26 Mar '24

26 Mar '24

Module: wine Branch: master Commit: 4e13c72f70edc057ac72e9ec86d83617f5e276d6 URL: https://gitlab.winehq.org/wine/wine/-/commit/4e13c72f70edc057ac72e9ec86d836… Author: Zebediah Figura <zfigura(a)codeweavers.com> Date: Fri Mar 15 16:32:14 2024 -0500 d3d11/tests: Test discarding a buffer in test_high_resource_count(). --- dlls/d3d11/tests/d3d11.c | 18 +++++++++++++++++- 1 file changed, 17 insertions(+), 1 deletion(-) diff --git a/dlls/d3d11/tests/d3d11.c b/dlls/d3d11/tests/d3d11.c index 221f17679b0..613dac57f5f 100644 --- a/dlls/d3d11/tests/d3d11.c +++ b/dlls/d3d11/tests/d3d11.c @@ -35619,15 +35619,18 @@ static void test_high_resource_count(void) D3D11_BUFFER_DESC buffer_desc = {0}; ID3D11ShaderResourceView *srvs[100]; ID3D11Texture2D *textures[50], *rt; + D3D11_MAPPED_SUBRESOURCE map_desc; ID3D11SamplerState *samplers[2]; ID3D11DeviceContext *context; ID3D11RenderTargetView *rtv; ID3D11Buffer *buffers[50]; ID3D11PixelShader *ps; ID3D11Device *device; + float *data_ptr; HRESULT hr; static const struct vec4 expect = {1274.0f, 637.0f, 1225.0f, 0.0f}; + static const struct vec4 expect2 = {1274.0f, 637.0f, 1325.0f, 0.0f}; static const DWORD ps_code[] = { @@ -36097,7 +36100,8 @@ static void test_high_resource_count(void) D3D11_SHADER_RESOURCE_VIEW_DESC srv_desc; buffer_desc.ByteWidth = sizeof(data); - buffer_desc.Usage = D3D11_USAGE_DEFAULT; + buffer_desc.Usage = D3D11_USAGE_DYNAMIC; + buffer_desc.CPUAccessFlags = D3D11_CPU_ACCESS_WRITE; buffer_desc.BindFlags = D3D11_BIND_SHADER_RESOURCE; hr = ID3D11Device_CreateBuffer(device, &buffer_desc, &data_desc, &buffers[i]); ok(hr == S_OK, "Got hr %#lx.\n", hr); @@ -36134,6 +36138,18 @@ static void test_high_resource_count(void) check_texture_vec4(rt, &expect, 0); + /* Discard the data in one of the buffers and draw again. */ + + hr = ID3D11DeviceContext_Map(context, (ID3D11Resource *)buffers[1], 0, D3D11_MAP_WRITE_DISCARD, 0, &map_desc); + ok(hr == S_OK, "Got hr %#lx.\n", hr); + data_ptr = map_desc.pData; + data_ptr[0] = 102.0f; + data_ptr[1] = 0.0f; + ID3D11DeviceContext_Unmap(context, (ID3D11Resource *)buffers[1], 0); + draw_quad(&test_context); + + todo_wine_if (!damavand) check_texture_vec4(rt, &expect2, 0); + ID3D11Texture2D_Release(rt); ID3D11RenderTargetView_Release(rtv); for (unsigned int i = 0; i < ARRAY_SIZE(srvs); ++i)

1 0

Zebediah Figura : wined3d: Recreate buffer textures when renaming the existing buffer storage.
by Alexandre Julliard 26 Mar '24

26 Mar '24

Module: wine Branch: master Commit: 58c680bacec189a46597d6f0b8f14f7061b423f2 URL: https://gitlab.winehq.org/wine/wine/-/commit/58c680bacec189a46597d6f0b8f14f… Author: Zebediah Figura <zfigura(a)codeweavers.com> Date: Fri Mar 15 16:33:18 2024 -0500 wined3d: Recreate buffer textures when renaming the existing buffer storage. Wine-Bug: https://bugs.winehq.org/show_bug.cgi?id=56438 --- dlls/d3d11/tests/d3d11.c | 2 +- dlls/wined3d/view.c | 10 ++++++++++ 2 files changed, 11 insertions(+), 1 deletion(-) diff --git a/dlls/d3d11/tests/d3d11.c b/dlls/d3d11/tests/d3d11.c index 613dac57f5f..24fdb32b829 100644 --- a/dlls/d3d11/tests/d3d11.c +++ b/dlls/d3d11/tests/d3d11.c @@ -36148,7 +36148,7 @@ static void test_high_resource_count(void) ID3D11DeviceContext_Unmap(context, (ID3D11Resource *)buffers[1], 0); draw_quad(&test_context); - todo_wine_if (!damavand) check_texture_vec4(rt, &expect2, 0); + check_texture_vec4(rt, &expect2, 0); ID3D11Texture2D_Release(rt); ID3D11RenderTargetView_Release(rtv); diff --git a/dlls/wined3d/view.c b/dlls/wined3d/view.c index a829a3b5ca2..09d1bfd32a1 100644 --- a/dlls/wined3d/view.c +++ b/dlls/wined3d/view.c @@ -269,7 +269,17 @@ static void create_buffer_texture(struct wined3d_gl_view *view, struct wined3d_c view->target = GL_TEXTURE_BUFFER; if (!view->name) + { gl_info->gl_ops.gl.p_glGenTextures(1, &view->name); + } + else if (gl_info->supported[ARB_BINDLESS_TEXTURE]) + { + /* If we already bound this view to a shader, we acquired a handle to + * it, and it's now immutable. This means we can't bind a new buffer + * storage to it, so recreate the texture. */ + gl_info->gl_ops.gl.p_glDeleteTextures(1, &view->name); + gl_info->gl_ops.gl.p_glGenTextures(1, &view->name); + } wined3d_context_gl_bind_texture(context_gl, GL_TEXTURE_BUFFER, view->name); if (gl_info->supported[ARB_TEXTURE_BUFFER_RANGE])

1 0

Zebediah Figura : wined3d: Skip bindless samplers with no uniform location.
by Alexandre Julliard 26 Mar '24

26 Mar '24

Module: wine Branch: master Commit: 37d71415edf7670ba5da70fc55c19c4e0f9b0463 URL: https://gitlab.winehq.org/wine/wine/-/commit/37d71415edf7670ba5da70fc55c19c… Author: Zebediah Figura <zfigura(a)codeweavers.com> Date: Fri Mar 15 16:34:03 2024 -0500 wined3d: Skip bindless samplers with no uniform location. This can happen legitimately if the sampler access was optimized out. --- dlls/wined3d/glsl_shader.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/dlls/wined3d/glsl_shader.c b/dlls/wined3d/glsl_shader.c index 42cdb585c50..933f134830a 100644 --- a/dlls/wined3d/glsl_shader.c +++ b/dlls/wined3d/glsl_shader.c @@ -10786,7 +10786,7 @@ static void shader_glsl_load_bindless_samplers(struct shader_glsl_priv *priv, st string_buffer_sprintf(sampler_name, "%s_sampler%u", prefix, entry->bind_idx); name_loc = GL_EXTCALL(glGetUniformLocation(ctx_data->glsl_program->id, sampler_name->buffer)); if (name_loc == -1) - ERR("No uniform location for shader %#x, binding %u, name %s.\n", shader_type, i, sampler_name->buffer); + continue; if ((view = state->shader_resource_view[shader_type][entry->resource_idx])) {

1 0

Tuomas Räsänen : winebus: Ignore udev events of devices which do not have devnodes.
by Alexandre Julliard 26 Mar '24

26 Mar '24

Module: wine Branch: master Commit: ef58b8c770023c13652954103b2a0b117e2deb94 URL: https://gitlab.winehq.org/wine/wine/-/commit/ef58b8c770023c13652954103b2a0b… Author: Tuomas Räsänen <tuomas.rasanen(a)opinsys.fi> Date: Fri Mar 22 11:41:25 2024 +0200 winebus: Ignore udev events of devices which do not have devnodes. Udev monitor monitors the whole input subsystem, but not all devices in the input subsystem have devnodes associated to them. This commit makes the event processing ignore such devices. All device handling assumes devices have devnodes, so here we just simply ignore all devices which do not have one. Previously, udev bus thread aborted when an event for a device without a devnode was processed: 10111.330:0068:0084:trace:hid:process_monitor_event Received action "remove" for udev device (null) 10111.330:0068:0084:warn:hid:bus_main_thread L"UDEV" bus wait returned status 0xc0000005 Just plugging in and out a normal mouse was enough cause this. This was because root input devices (which do not have devnodes) were handled too and find_device_from_devnode() choked on NULL argument. --- dlls/winebus.sys/bus_udev.c | 28 ++++++++++++++++++++-------- 1 file changed, 20 insertions(+), 8 deletions(-) diff --git a/dlls/winebus.sys/bus_udev.c b/dlls/winebus.sys/bus_udev.c index 419de2e7af3..2cd995a4178 100644 --- a/dlls/winebus.sys/bus_udev.c +++ b/dlls/winebus.sys/bus_udev.c @@ -1710,26 +1710,38 @@ static void process_monitor_event(struct udev_monitor *monitor) { struct base_device *impl; struct udev_device *dev; - const char *action; + const char *action, *devnode, *syspath; dev = udev_monitor_receive_device(monitor); if (!dev) { - FIXME("Failed to get device that has changed\n"); + ERR("Failed to get device that has changed\n"); return; } action = udev_device_get_action(dev); - TRACE("Received action %s for udev device %s\n", debugstr_a(action), - debugstr_a(udev_device_get_devnode(dev))); - - if (!action) - WARN("No action received\n"); + syspath = udev_device_get_syspath(dev); + devnode = udev_device_get_devnode(dev); + TRACE("Received action %s for udev device %s (%p) devnode %s\n", + debugstr_a(action), debugstr_a(syspath), dev, debugstr_a(devnode)); + + if (!syspath) + ERR("udev device %p does not have syspath!\n", dev); + else if (!action) + ERR("event for udev device %s does not have any action!\n", syspath); + else if (!devnode) + { + /* Pretty normal case, not all devices have associated + * devnodes. For example root input devices do not, but + * related/child mouse and event devices do. + */ + TRACE("udev device %s does not have devnode, ignoring\n", syspath); + } else if (strcmp(action, "remove")) udev_add_device(dev, -1); else { - impl = find_device_from_devnode(udev_device_get_devnode(dev)); + impl = find_device_from_devnode(devnode); if (impl) bus_event_queue_device_removed(&event_queue, &impl->unix_device); else WARN("failed to find device for udev device %p\n", dev); }

1 0

Eric Pouech : evr: Remove useless code.
by Alexandre Julliard 26 Mar '24

26 Mar '24

Module: wine Branch: master Commit: d950e1b61f34e8bc333ec558edcbfc46bfbb9936 URL: https://gitlab.winehq.org/wine/wine/-/commit/d950e1b61f34e8bc333ec558edcbfc… Author: Eric Pouech <epouech(a)codeweavers.com> Date: Wed Mar 13 09:22:58 2024 +0100 evr: Remove useless code. Signed-off-by: Eric Pouech <epouech(a)codeweavers.com> --- dlls/evr/evr.c | 1 - 1 file changed, 1 deletion(-) diff --git a/dlls/evr/evr.c b/dlls/evr/evr.c index 52c5de5e2c4..362c4f1e76f 100644 --- a/dlls/evr/evr.c +++ b/dlls/evr/evr.c @@ -409,7 +409,6 @@ static HRESULT evr_copy_sample_buffer(struct evr *filter, const GUID *subtype, I /* All other formats are 32-bit, single plane. */ MFCopyImage(locked_rect.pBits, locked_rect.Pitch, src, src_stride, width * 4, lines); } - MFCopyImage(locked_rect.pBits, locked_rect.Pitch, src, src_stride, width, lines); IDirect3DSurface9_UnlockRect(surface); }

1 0

Paul Gofman : ntdll: Support generic xstate in Unix-side manipulation functions.
by Alexandre Julliard 26 Mar '24

26 Mar '24

Module: wine Branch: master Commit: 6bc2c092327bae6cbf73b75fd33581ee5140a285 URL: https://gitlab.winehq.org/wine/wine/-/commit/6bc2c092327bae6cbf73b75fd33581… Author: Paul Gofman <pgofman(a)codeweavers.com> Date: Wed Jan 17 16:33:33 2024 -0600 ntdll: Support generic xstate in Unix-side manipulation functions. --- dlls/ntdll/unix/signal_i386.c | 2 ++ dlls/ntdll/unix/signal_x86_64.c | 4 ++++ dlls/ntdll/unix/system.c | 48 +++++++++++++++++++++++++++++++++++++---- 3 files changed, 50 insertions(+), 4 deletions(-) diff --git a/dlls/ntdll/unix/signal_i386.c b/dlls/ntdll/unix/signal_i386.c index 94799522202..61f42806f5d 100644 --- a/dlls/ntdll/unix/signal_i386.c +++ b/dlls/ntdll/unix/signal_i386.c @@ -1139,6 +1139,8 @@ NTSTATUS WINAPI NtGetContextThread( HANDLE handle, CONTEXT *context ) if (context_ex->XState.Length < xstate_get_size( xstate->CompactionMask, xstate->Mask )) return STATUS_BUFFER_OVERFLOW; copy_xstate( xstate, &frame->xstate, xstate->Mask ); + /* copy_xstate may use avx in memcpy, restore xstate not to break the tests. */ + frame->restore_flags |= CONTEXT_XSTATE; } } /* update the cached version of the debug registers */ diff --git a/dlls/ntdll/unix/signal_x86_64.c b/dlls/ntdll/unix/signal_x86_64.c index f4a6fcce8ca..d8d16917942 100644 --- a/dlls/ntdll/unix/signal_x86_64.c +++ b/dlls/ntdll/unix/signal_x86_64.c @@ -1172,6 +1172,8 @@ NTSTATUS WINAPI NtGetContextThread( HANDLE handle, CONTEXT *context ) if (context_ex->XState.Length < xstate_get_size( xstate->CompactionMask, xstate->Mask )) return STATUS_BUFFER_OVERFLOW; copy_xstate( xstate, &frame->xstate, xstate->Mask ); + /* copy_xstate may use avx in memcpy, restore xstate not to break the tests. */ + frame->restore_flags |= CONTEXT_XSTATE; } } /* update the cached version of the debug registers */ @@ -1382,6 +1384,8 @@ NTSTATUS get_thread_wow64_context( HANDLE handle, void *ctx, ULONG size ) if (context_ex->XState.Length < xstate_get_size( xstate->CompactionMask, xstate->Mask )) return STATUS_BUFFER_OVERFLOW; copy_xstate( xstate, &frame->xstate, xstate->Mask ); + /* copy_xstate may use avx in memcpy, restore xstate not to break the tests. */ + frame->restore_flags |= CONTEXT_XSTATE; } } return STATUS_SUCCESS; diff --git a/dlls/ntdll/unix/system.c b/dlls/ntdll/unix/system.c index 4c6c4cd23e2..068eb9d6839 100644 --- a/dlls/ntdll/unix/system.c +++ b/dlls/ntdll/unix/system.c @@ -250,20 +250,60 @@ BOOL xstate_compaction_enabled = FALSE; UINT64 xstate_supported_features_mask; UINT64 xstate_features_size; +static int xstate_feature_offset[64] = {0, 0, 576}; +static int xstate_feature_size[64] = {0, 0, 256}; +static UINT64 xstate_aligned_features; + +static int next_xstate_offset( int off, UINT64 compaction_mask, int feature_idx ) +{ + const UINT64 feature_mask = (UINT64)1 << feature_idx; + + if (!compaction_mask) return xstate_feature_offset[feature_idx + 1] - sizeof(XSAVE_FORMAT); + + if (compaction_mask & feature_mask) off += xstate_feature_size[feature_idx]; + if (xstate_aligned_features & (feature_mask << 1)) + off = (off + 63) & ~63; + return off; +} + unsigned int xstate_get_size( UINT64 compaction_mask, UINT64 mask ) { - if (!(mask & ((UINT64)1 << XSTATE_AVX))) return sizeof(XSAVE_AREA_HEADER); - return sizeof(XSAVE_AREA_HEADER) + sizeof(YMMCONTEXT); + unsigned int i; + int off; + + mask >>= 2; + off = sizeof(XSAVE_AREA_HEADER); + i = 2; + while (mask) + { + if (mask == 1) return off + xstate_feature_size[i]; + off = next_xstate_offset( off, compaction_mask, i ); + mask >>= 1; + ++i; + } + return off; } void copy_xstate( XSAVE_AREA_HEADER *dst, XSAVE_AREA_HEADER *src, UINT64 mask ) { + unsigned int i; + int src_off, dst_off; + mask &= xstate_extended_features() & src->Mask; if (src->CompactionMask) mask &= src->CompactionMask; if (dst->CompactionMask) mask &= dst->CompactionMask; dst->Mask = (dst->Mask & ~xstate_extended_features()) | mask; - if (mask & ((UINT64)1 << XSTATE_AVX)) - *(YMMCONTEXT *)(dst + 1) = *(YMMCONTEXT *)(src + 1); + mask >>= 2; + src_off = dst_off = sizeof(XSAVE_AREA_HEADER); + i = 2; + while (1) + { + if (mask & 1) memcpy( (char *)dst + dst_off, (char *)src + src_off, xstate_feature_size[i] ); + if (!(mask >>= 1)) break; + src_off = next_xstate_offset( src_off, src->CompactionMask, i ); + dst_off = next_xstate_offset( dst_off, dst->CompactionMask, i ); + ++i; + } } #define AUTH 0x68747541 /* "Auth" */

1 0

Paul Gofman : ntdll: Factor out xstate_from_server() function.
by Alexandre Julliard 26 Mar '24

26 Mar '24

Module: wine Branch: master Commit: 25e0a2580758282d679dafa557b2c1771410e970 URL: https://gitlab.winehq.org/wine/wine/-/commit/25e0a2580758282d679dafa557b2c1… Author: Paul Gofman <pgofman(a)codeweavers.com> Date: Mon Mar 18 15:00:06 2024 -0600 ntdll: Factor out xstate_from_server() function. --- dlls/ntdll/unix/thread.c | 82 +++++++++++++++--------------------------------- 1 file changed, 26 insertions(+), 56 deletions(-) diff --git a/dlls/ntdll/unix/thread.c b/dlls/ntdll/unix/thread.c index 60c833b0320..0d98473f01b 100644 --- a/dlls/ntdll/unix/thread.c +++ b/dlls/ntdll/unix/thread.c @@ -613,6 +613,28 @@ static NTSTATUS context_to_server( context_t *to, USHORT to_machine, const void } +/*********************************************************************** + * xstate_from_server + * + * Copy xstate from the server format. + */ +static void xstate_from_server( CONTEXT_EX *xctx, const context_t *from ) +{ + XSTATE *xs = (XSTATE *)((char *)xctx + xctx->XState.Offset); + unsigned int i; + + xs->Mask &= ~4; + if (xs->CompactionMask) xs->CompactionMask = 0x8000000000000004; + for (i = 0; i < ARRAY_SIZE( from->ymm.regs.ymm_high); i++) + { + if (!from->ymm.regs.ymm_high[i].low && !from->ymm.regs.ymm_high[i].high) continue; + memcpy( &xs->YmmContext, &from->ymm.regs, sizeof(xs->YmmContext) ); + xs->Mask |= 4; + break; + } +} + + /*********************************************************************** * context_from_server * @@ -686,20 +708,7 @@ static NTSTATUS context_from_server( void *dst, const context_t *from, USHORT ma memcpy( to->ExtendedRegisters, from->ext.i386_regs, sizeof(to->ExtendedRegisters) ); } if ((from->flags & SERVER_CTX_YMM_REGISTERS) && (to_flags & CONTEXT_I386_XSTATE)) - { - CONTEXT_EX *xctx = (CONTEXT_EX *)(to + 1); - XSTATE *xs = (XSTATE *)((char *)xctx + xctx->XState.Offset); - - xs->Mask &= ~4; - if (xs->CompactionMask) xs->CompactionMask = 0x8000000000000004; - for (i = 0; i < ARRAY_SIZE( from->ymm.regs.ymm_high); i++) - { - if (!from->ymm.regs.ymm_high[i].low && !from->ymm.regs.ymm_high[i].high) continue; - memcpy( &xs->YmmContext, &from->ymm.regs, sizeof(xs->YmmContext) ); - xs->Mask |= 4; - break; - } - } + xstate_from_server( (CONTEXT_EX *)(to + 1), from ); return STATUS_SUCCESS; } @@ -760,20 +769,7 @@ static NTSTATUS context_from_server( void *dst, const context_t *from, USHORT ma to->Dr7 = from->debug.x86_64_regs.dr7; } if ((from->flags & SERVER_CTX_YMM_REGISTERS) && (to_flags & CONTEXT_I386_XSTATE)) - { - CONTEXT_EX *xctx = (CONTEXT_EX *)(to + 1); - XSTATE *xs = (XSTATE *)((char *)xctx + xctx->XState.Offset); - - xs->Mask &= ~4; - if (xs->CompactionMask) xs->CompactionMask = 0x8000000000000004; - for (i = 0; i < ARRAY_SIZE( from->ymm.regs.ymm_high); i++) - { - if (!from->ymm.regs.ymm_high[i].low && !from->ymm.regs.ymm_high[i].high) continue; - memcpy( &xs->YmmContext, &from->ymm.regs, sizeof(xs->YmmContext) ); - xs->Mask |= 4; - break; - } - } + xstate_from_server( (CONTEXT_EX *)(to + 1), from ); return STATUS_SUCCESS; } @@ -835,20 +831,7 @@ static NTSTATUS context_from_server( void *dst, const context_t *from, USHORT ma to->Dr7 = from->debug.x86_64_regs.dr7; } if ((from->flags & SERVER_CTX_YMM_REGISTERS) && (to_flags & CONTEXT_AMD64_XSTATE)) - { - CONTEXT_EX *xctx = (CONTEXT_EX *)(to + 1); - XSTATE *xs = (XSTATE *)((char *)xctx + xctx->XState.Offset); - - xs->Mask &= ~4; - if (xs->CompactionMask) xs->CompactionMask = 0x8000000000000004; - for (i = 0; i < ARRAY_SIZE( from->ymm.regs.ymm_high); i++) - { - if (!from->ymm.regs.ymm_high[i].low && !from->ymm.regs.ymm_high[i].high) continue; - memcpy( &xs->YmmContext, &from->ymm.regs, sizeof(xs->YmmContext) ); - xs->Mask |= 4; - break; - } - } + xstate_from_server( (CONTEXT_EX *)(to + 1), from ); return STATUS_SUCCESS; } @@ -917,20 +900,7 @@ static NTSTATUS context_from_server( void *dst, const context_t *from, USHORT ma to->Dr7 = from->debug.i386_regs.dr7; } if ((from->flags & SERVER_CTX_YMM_REGISTERS) && (to_flags & CONTEXT_AMD64_XSTATE)) - { - CONTEXT_EX *xctx = (CONTEXT_EX *)(to + 1); - XSTATE *xs = (XSTATE *)((char *)xctx + xctx->XState.Offset); - - xs->Mask &= ~4; - if (xs->CompactionMask) xs->CompactionMask = 0x8000000000000004; - for (i = 0; i < ARRAY_SIZE( from->ymm.regs.ymm_high); i++) - { - if (!from->ymm.regs.ymm_high[i].low && !from->ymm.regs.ymm_high[i].high) continue; - memcpy( &xs->YmmContext, &from->ymm.regs, sizeof(xs->YmmContext) ); - xs->Mask |= 4; - break; - } - } + xstate_from_server( (CONTEXT_EX *)(to + 1), from ); return STATUS_SUCCESS; }

1 0

Paul Gofman : ntdll: Factor out context_to_server() function.
by Alexandre Julliard 26 Mar '24

26 Mar '24

Module: wine Branch: master Commit: a9d77d098928f6521bef7c1b1739429b8fe11ebd URL: https://gitlab.winehq.org/wine/wine/-/commit/a9d77d098928f6521bef7c1b173942… Author: Paul Gofman <pgofman(a)codeweavers.com> Date: Mon Mar 18 15:29:50 2024 -0600 ntdll: Factor out context_to_server() function. --- dlls/ntdll/unix/thread.c | 46 ++++++++++++++++++---------------------------- 1 file changed, 18 insertions(+), 28 deletions(-) diff --git a/dlls/ntdll/unix/thread.c b/dlls/ntdll/unix/thread.c index 0d98473f01b..800f2cc1968 100644 --- a/dlls/ntdll/unix/thread.c +++ b/dlls/ntdll/unix/thread.c @@ -227,6 +227,20 @@ static unsigned int get_native_context_flags( USHORT native_machine, USHORT wow_ } +/*********************************************************************** + * xstate_to_server + * + * Copy xstate to the server format. + */ +static void xstate_to_server( context_t *to, const CONTEXT_EX *xctx ) +{ + const XSTATE *xs = (const XSTATE *)((const char *)xctx + xctx->XState.Offset); + + to->flags |= SERVER_CTX_YMM_REGISTERS; + if (xs->Mask & 4) memcpy( &to->ymm.regs.ymm_high, &xs->YmmContext, sizeof(xs->YmmContext) ); +} + + /*********************************************************************** * context_to_server * @@ -303,13 +317,7 @@ static NTSTATUS context_to_server( context_t *to, USHORT to_machine, const void memcpy( to->ext.i386_regs, from->ExtendedRegisters, sizeof(to->ext.i386_regs) ); } if (flags & CONTEXT_I386_XSTATE) - { - const CONTEXT_EX *xctx = (const CONTEXT_EX *)(from + 1); - const XSTATE *xs = (const XSTATE *)((const char *)xctx + xctx->XState.Offset); - - to->flags |= SERVER_CTX_YMM_REGISTERS; - if (xs->Mask & 4) memcpy( &to->ymm.regs.ymm_high, &xs->YmmContext, sizeof(xs->YmmContext) ); - } + xstate_to_server( to, (const CONTEXT_EX *)(from + 1) ); return STATUS_SUCCESS; } @@ -367,13 +375,7 @@ static NTSTATUS context_to_server( context_t *to, USHORT to_machine, const void fpu_to_fpux( (XMM_SAVE_AREA32 *)to->fp.x86_64_regs.fpregs, &from->FloatSave ); } if (flags & CONTEXT_I386_XSTATE) - { - const CONTEXT_EX *xctx = (const CONTEXT_EX *)(from + 1); - const XSTATE *xs = (const XSTATE *)((const char *)xctx + xctx->XState.Offset); - - to->flags |= SERVER_CTX_YMM_REGISTERS; - if (xs->Mask & 4) memcpy( &to->ymm.regs.ymm_high, &xs->YmmContext, sizeof(xs->YmmContext) ); - } + xstate_to_server( to, (const CONTEXT_EX *)(from + 1) ); return STATUS_SUCCESS; } @@ -434,13 +436,7 @@ static NTSTATUS context_to_server( context_t *to, USHORT to_machine, const void to->debug.x86_64_regs.dr7 = from->Dr7; } if (flags & CONTEXT_AMD64_XSTATE) - { - const CONTEXT_EX *xctx = (const CONTEXT_EX *)(from + 1); - const XSTATE *xs = (const XSTATE *)((const char *)xctx + xctx->XState.Offset); - - to->flags |= SERVER_CTX_YMM_REGISTERS; - if (xs->Mask & 4) memcpy( &to->ymm.regs.ymm_high, &xs->YmmContext, sizeof(xs->YmmContext) ); - } + xstate_to_server( to, (const CONTEXT_EX *)(from + 1) ); return STATUS_SUCCESS; } @@ -505,13 +501,7 @@ static NTSTATUS context_to_server( context_t *to, USHORT to_machine, const void to->debug.i386_regs.dr7 = from->Dr7; } if (flags & CONTEXT_AMD64_XSTATE) - { - const CONTEXT_EX *xctx = (const CONTEXT_EX *)(from + 1); - const XSTATE *xs = (const XSTATE *)((const char *)xctx + xctx->XState.Offset); - - to->flags |= SERVER_CTX_YMM_REGISTERS; - if (xs->Mask & 4) memcpy( &to->ymm.regs.ymm_high, &xs->YmmContext, sizeof(xs->YmmContext) ); - } + xstate_to_server( to, (const CONTEXT_EX *)(from + 1) ); return STATUS_SUCCESS; }

1 0

Paul Gofman : ntdll: Mind generic xstate masks in server context conversion.
by Alexandre Julliard 26 Mar '24

26 Mar '24

Module: wine Branch: master Commit: 3dea242677be1ca7f18ca95b6d9d0aacc5c3f1b9 URL: https://gitlab.winehq.org/wine/wine/-/commit/3dea242677be1ca7f18ca95b6d9d0a… Author: Paul Gofman <pgofman(a)codeweavers.com> Date: Mon Mar 18 15:14:50 2024 -0600 ntdll: Mind generic xstate masks in server context conversion. --- dlls/ntdll/unix/thread.c | 11 +++++++++-- 1 file changed, 9 insertions(+), 2 deletions(-) diff --git a/dlls/ntdll/unix/thread.c b/dlls/ntdll/unix/thread.c index 800f2cc1968..7ae5c022fb2 100644 --- a/dlls/ntdll/unix/thread.c +++ b/dlls/ntdll/unix/thread.c @@ -236,6 +236,7 @@ static void xstate_to_server( context_t *to, const CONTEXT_EX *xctx ) { const XSTATE *xs = (const XSTATE *)((const char *)xctx + xctx->XState.Offset); + if (xs->CompactionMask && !(xs->CompactionMask & 4)) return; to->flags |= SERVER_CTX_YMM_REGISTERS; if (xs->Mask & 4) memcpy( &to->ymm.regs.ymm_high, &xs->YmmContext, sizeof(xs->YmmContext) ); } @@ -613,8 +614,14 @@ static void xstate_from_server( CONTEXT_EX *xctx, const context_t *from ) XSTATE *xs = (XSTATE *)((char *)xctx + xctx->XState.Offset); unsigned int i; - xs->Mask &= ~4; - if (xs->CompactionMask) xs->CompactionMask = 0x8000000000000004; + xs->Mask &= ~(ULONG64)4; + + if (xs->CompactionMask) + { + xs->CompactionMask &= ~(UINT64)3; + if (!(xs->CompactionMask & 4)) return; + } + for (i = 0; i < ARRAY_SIZE( from->ymm.regs.ymm_high); i++) { if (!from->ymm.regs.ymm_high[i].low && !from->ymm.regs.ymm_high[i].high) continue;

1 0

Paul Gofman : ntdll: Support more xstate features.
by Alexandre Julliard 26 Mar '24

26 Mar '24

Module: wine Branch: master Commit: 29c73ee17335b30f3f49c8b3562742c6a35b482c URL: https://gitlab.winehq.org/wine/wine/-/commit/29c73ee17335b30f3f49c8b3562742… Author: Paul Gofman <pgofman(a)codeweavers.com> Date: Thu Jan 18 12:59:26 2024 -0600 ntdll: Support more xstate features. --- dlls/ntdll/tests/exception.c | 2 +- dlls/ntdll/unix/system.c | 40 ++++++++++++++++++++++++++---- programs/wineboot/wineboot.c | 58 ++++++++++++++++++++++++++++++++------------ 3 files changed, 79 insertions(+), 21 deletions(-) diff --git a/dlls/ntdll/tests/exception.c b/dlls/ntdll/tests/exception.c index 67b02cbd8b6..7a4e04ec5e0 100644 --- a/dlls/ntdll/tests/exception.c +++ b/dlls/ntdll/tests/exception.c @@ -9171,7 +9171,7 @@ static DWORD test_extended_context_handler(EXCEPTION_RECORD *rec, EXCEPTION_REGI } else { - ok(xs->Mask == (xsaveopt_enabled ? 0 : 4), "Got unexpected Mask %#I64x.\n", xs->Mask); + ok((xs->Mask & 7) == (xsaveopt_enabled ? 0 : 4), "Got unexpected Mask %#I64x.\n", xs->Mask); /* The save area has garbage if xsaveopt is available, so we can't test * its contents. */ diff --git a/dlls/ntdll/unix/system.c b/dlls/ntdll/unix/system.c index 068eb9d6839..15c4a49e611 100644 --- a/dlls/ntdll/unix/system.c +++ b/dlls/ntdll/unix/system.c @@ -250,8 +250,8 @@ BOOL xstate_compaction_enabled = FALSE; UINT64 xstate_supported_features_mask; UINT64 xstate_features_size; -static int xstate_feature_offset[64] = {0, 0, 576}; -static int xstate_feature_size[64] = {0, 0, 256}; +static int xstate_feature_offset[64]; +static int xstate_feature_size[64]; static UINT64 xstate_aligned_features; static int next_xstate_offset( int off, UINT64 compaction_mask, int feature_idx ) @@ -345,6 +345,21 @@ __ASM_GLOBAL_FUNC( do_cpuid, "ret" ) #endif +extern UINT64 do_xgetbv( unsigned int cx); +#ifdef __i386__ +__ASM_GLOBAL_FUNC( do_xgetbv, + "movl 4(%esp),%ecx\n\t" + "xgetbv\n\t" + "ret" ) +#else +__ASM_GLOBAL_FUNC( do_xgetbv, + "movl %edi,%ecx\n\t" + "xgetbv\n\t" + "shlq $32,%rdx\n\t" + "orq %rdx,%rax\n\t" + "ret" ) +#endif + #ifdef __i386__ extern int have_cpuid(void); __ASM_GLOBAL_FUNC( have_cpuid, @@ -403,8 +418,11 @@ static void get_cpuid_name( char *buffer ) static void get_cpuinfo( SYSTEM_CPU_INFORMATION *info ) { + static const ULONG64 wine_xstate_supported_features = 0xff; /* XSTATE_AVX, XSTATE_MPX_BNDREGS, XSTATE_MPX_BNDCSR, + * XSTATE_AVX512_KMASK, XSTATE_AVX512_ZMM_H, XSTATE_AVX512_ZMM */ unsigned int regs[4], regs2[4], regs3[4]; ULONGLONG features; + unsigned int i; #if defined(__i386__) info->ProcessorArchitecture = PROCESSOR_ARCHITECTURE_INTEL; @@ -454,13 +472,25 @@ static void get_cpuinfo( SYSTEM_CPU_INFORMATION *info ) { do_cpuid( 0x0000000d, 1, regs3 ); /* get XSAVE details */ if (regs3[0] & 2) xstate_compaction_enabled = TRUE; - xstate_supported_features_mask = 3; - if (features & CPU_FEATURE_AVX) - xstate_supported_features_mask |= (UINT64)1 << XSTATE_AVX; + + do_cpuid( 0x0000000d, 0, regs3 ); /* get user xstate features */ + xstate_supported_features_mask = ((ULONG64)regs3[3] << 32) | regs3[0]; + xstate_supported_features_mask &= do_xgetbv( 0 ) & wine_xstate_supported_features; + TRACE("xstate_supported_features_mask %#llx.\n", (long long)xstate_supported_features_mask); + for (i = 2; i < 64; ++i) + { + if (!(xstate_supported_features_mask & ((ULONG64)1 << i))) continue; + do_cpuid( 0x0000000d, i, regs3 ); /* get user xstate features */ + xstate_feature_offset[i] = regs3[1]; + xstate_feature_size[i] = regs3[0]; + if (regs3[2] & 2) xstate_aligned_features |= (ULONG64)1 << i; + TRACE("xstate[%d] offset %d, size %d, aligned %d.\n", i, xstate_feature_offset[i], xstate_feature_size[i], !!(regs3[2] & 2)); + } xstate_features_size = xstate_get_size( xstate_compaction_enabled ? 0x8000000000000000 | xstate_supported_features_mask : 0, xstate_supported_features_mask ) - sizeof(XSAVE_AREA_HEADER); xstate_features_size = (xstate_features_size + 15) & ~15; + TRACE("xstate_features_size %lld.\n", (long long)xstate_features_size); } if (regs[1] == AUTH && regs[3] == ENTI && regs[2] == CAMD) diff --git a/programs/wineboot/wineboot.c b/programs/wineboot/wineboot.c index 1c1ad858fa4..67d9508389f 100644 --- a/programs/wineboot/wineboot.c +++ b/programs/wineboot/wineboot.c @@ -193,10 +193,27 @@ static DWORD set_reg_value_dword( HKEY hkey, const WCHAR *name, DWORD value ) #if defined(__i386__) || defined(__x86_64__) +extern UINT64 WINAPI do_xgetbv( unsigned int cx); +#ifdef __i386__ +__ASM_STDCALL_FUNC( do_xgetbv, 4, + "movl 4(%esp),%ecx\n\t" + "xgetbv\n\t" + "ret $4" ) +#else +__ASM_GLOBAL_FUNC( do_xgetbv, + "xgetbv\n\t" + "shlq $32,%rdx\n\t" + "orq %rdx,%rax\n\t" + "ret" ) +#endif + static void initialize_xstate_features(struct _KUSER_SHARED_DATA *data) { + static const ULONG64 wine_xstate_supported_features = 0xfc; /* XSTATE_AVX, XSTATE_MPX_BNDREGS, XSTATE_MPX_BNDCSR, + * XSTATE_AVX512_KMASK, XSTATE_AVX512_ZMM_H, XSTATE_AVX512_ZMM */ XSTATE_CONFIGURATION *xstate = &data->XState; - unsigned int i; + ULONG64 supported_mask; + unsigned int i, off; int regs[4]; if (!data->ProcessorFeatures[PF_AVX_INSTRUCTIONS_AVAILABLE]) @@ -215,29 +232,40 @@ static void initialize_xstate_features(struct _KUSER_SHARED_DATA *data) __cpuidex(regs, 0xd, 0); TRACE("XSAVE details %#x, %#x, %#x, %#x.\n", regs[0], regs[1], regs[2], regs[3]); - if (!(regs[0] & XSTATE_AVX)) + supported_mask = ((ULONG64)regs[3] << 32) | regs[0]; + supported_mask &= do_xgetbv(0) & wine_xstate_supported_features; + if (!(supported_mask >> 2)) return; - xstate->EnabledFeatures = (1 << XSTATE_LEGACY_FLOATING_POINT) | (1 << XSTATE_LEGACY_SSE) | (1 << XSTATE_AVX); + xstate->EnabledFeatures = (1 << XSTATE_LEGACY_FLOATING_POINT) | (1 << XSTATE_LEGACY_SSE) | supported_mask; xstate->EnabledVolatileFeatures = xstate->EnabledFeatures; - xstate->Size = sizeof(XSAVE_FORMAT) + sizeof(XSTATE); xstate->AllFeatureSize = regs[1]; - xstate->AllFeatures[0] = offsetof(XSAVE_FORMAT, XmmRegisters); - xstate->AllFeatures[1] = sizeof(M128A) * 16; - xstate->AllFeatures[2] = sizeof(YMMCONTEXT); - - for (i = 0; i < 3; ++i) - xstate->Features[i].Size = xstate->AllFeatures[i]; - - xstate->Features[1].Offset = xstate->Features[0].Size; - xstate->Features[2].Offset = sizeof(XSAVE_FORMAT) + offsetof(XSTATE, YmmContext); __cpuidex(regs, 0xd, 1); xstate->OptimizedSave = regs[0] & 1; xstate->CompactionEnabled = !!(regs[0] & 2); - __cpuidex(regs, 0xd, 2); - TRACE("XSAVE feature 2 %#x, %#x, %#x, %#x.\n", regs[0], regs[1], regs[2], regs[3]); + xstate->Features[0].Size = xstate->AllFeatures[0] = offsetof(XSAVE_FORMAT, XmmRegisters); + xstate->Features[1].Size = xstate->AllFeatures[1] = sizeof(M128A) * 16; + xstate->Features[1].Offset = xstate->Features[0].Size; + off = sizeof(XSAVE_FORMAT) + sizeof(XSAVE_AREA_HEADER); + supported_mask >>= 2; + for (i = 2; supported_mask; ++i, supported_mask >>= 1) + { + if (!(supported_mask & 1)) continue; + __cpuidex( regs, 0xd, i ); + xstate->Features[i].Offset = regs[1]; + xstate->Features[i].Size = xstate->AllFeatures[i] = regs[0]; + if (regs[2] & 2) + { + xstate->AlignedFeatures |= (ULONG64)1 << i; + off = (off + 63) & ~63; + } + off += xstate->Features[i].Size; + TRACE("xstate[%d] offset %lu, size %lu, aligned %d.\n", i, xstate->Features[i].Offset, xstate->Features[i].Size, !!(regs[2] & 2)); + } + xstate->Size = xstate->CompactionEnabled ? off : xstate->Features[i - 1].Offset + xstate->Features[i - 1].Size; + TRACE("xstate size %lu, compacted %d, optimized %d.\n", xstate->Size, xstate->CompactionEnabled, xstate->OptimizedSave); } static BOOL is_tsc_trusted_by_the_kernel(void)

1 0